this is unironically the right way to build a startup

After a million years I finally uploaded a new video. This time about RoPE! Let me know what you think! youtu.be/V8r__fXx7tU?si…

Any RLHF experts here? Currently working on a video about PPO / GRPO etc., and I want to introduce importance sampling (see 1st screenshot). Is this a valid step to take, to go from the expected value over trajectories to expected value over actions? And is my explanation for

This is Ray3. The world’s first reasoning video model, and the first to generate studio-grade HDR. Now with an all-new Draft Mode for rapid iteration in creative workflows, and state of the art physics and consistency. Available now for free in Dream Machine.

It took an incredible amount of energy to get here, but now we're ready to unleash Ray3, our new frontier video model with reasoning capabilities. I especially love the HDR video generations, the colors and lighting just pop in ways that make SDR look dull. Check it out!

4 year pytorch bug where all reduce operation produces INCORRECT gradients with no warning. Still not patched. Initially reported by Jim Fan. Sharing this in case anyone is having mysterious gradient explosions. github.com/pytorch/pytorc…

Changing the autoencoder in latent diffusion models is easier than you think. 🚀 Introducing DC-Gen – a post-training acceleration framework that works with any pre-trained diffusion model, boosting efficiency by transferring it into a deeply compressed latent space with

We release DC-VideoGen, a new post-training framework for accelerating video diffusion models. Key features: 🎬 Supports video generation up to 2160×3840 (4K) resolution on a single H100 GPU ⚡ Delivers 14.8× faster inference than the base model while achieving comparable or

much more convinced after getting my own results: LoRA with rank=1 learns (and generalizes) as well as full-tuning while saving 43% vRAM usage! allows me to RL bigger models with limited resources😆 script: github.com/sail-sg/oat/bl…

Announcing OpenMoE 2, the first-ever architectural study of sparse diffusion language models, trained from scratch. ✅ Expert-choice MoE × diffusion ✅ Ultra-wide FLOPs/param range (sparse → super-dense) ✅ Perfect load-balance (no aux loss) ✅ +20% throughput ✅ adaptive

Imagine a world where you can build any workflow. A new Runway is coming soon.

After 1.5 years at Luma AI, I have decided to leave and that the AI field is not my thing anymore and that I will pursue a career as a professional surfer. Just kidding, I‘m joining Runway

This is my personal “Zuck poached top talent” moment for AI video Dome is exceptional. Was a pleasure working with him at Luma and excited to see where his future leads.

Today we’re releasing another new batch of Runway Apps with a focus on VFX. The first is the Change Weather app. Now you can upload a video and simply describe what conditions you’d like to see. Make a sunny day look overcast or bring torrential downpours. (1/5)

today we're open-sourcing Krea Realtime. this 14B autoregressive model is 10x larger than any open-source equivalent, and it can generate long-form videos at 11 fps on a single B200. weights and technical report below 👇

An easier way to edit your videos. The Remove from Video app. Simply upload a video, describe what element you’d like removed and that’s it. No rotoscoping, no complex prompting. Just describe it. Try it now at the link below.

Apps have become my default mode for making things. They make it so easy to get exactly what you want without any complications.

Today Runway turned 7 years old. What was impossible became inevitable. Happy birthday, Runway. Grateful to everyone who made this possible. The next 7 years start now.

Serious take: solid solutions to (a) long context and (b) continuous latent reasoning is what is missing for "real" AI.

there’s ai slop and there’s ai kino. this is ai kino.